The objective of this study was to develop and validate a method for identifying Medicare beneficiaries with diabetes by using Medicare claims data. We used self-reports of diabetes status from participants in the Medicare Current Beneficiary Survey to determine disease status, and then we examined these participants' Medicare claims. Using self-reported diabetes status as the "gold standard," we determined the sensitivity, specificity, and reliability of claims data in identifying beneficiaries with diabetes. We found that to construct a method that is adequately sensitive (> or = 70%), highly specific (> or = 97.5%), and reliable (kappa > or = 0.80), researchers must combine information from different types of Medicare claims files, use 2 years of data to identify cases, and require at least 2 diagnoses of diabetes among claims involving ambulatory care. Since these criteria are met by more than one method, the choice of method should be governed by the goals of the research as well as more practical concerns.